Hello, everyone. This is ToTheMoon with technology news and insight from Silicon Valley and around the world. Perhaps no subject has been exploited more heavily throughout the rise of AI than the future of work: what people should study, which professions will survive, and which will disappear. An extraordinary amount of marketing has been built around that question. I want to examine it through programming. A large part of our audience is not made up of programmers or IT specialists.
For many people, those who work in software seem almost extraterrestrial—set apart as a separate class. One reason is obvious: compared with many other professions, software professionals gained access to unusually high incomes. Programming is nevertheless the clearest field through which to understand what AI may do to work. I follow people from very different backgrounds because their perspectives reveal different parts of the picture. When Sam Altman of OpenAI, Dario Amodei of Anthropic, or Demis Hassabis of Google DeepMind speaks, we hear the people building the systems.
It is also useful to listen to business leaders, broader visionaries, working programmers, politicians, economists, cultural figures, and people associated with AI companies in less direct ways. I recently saw a statement from someone who is either the CEO or a board leader at Yandex—I do not know his exact current title, and the individual is not the important part. What matters is the company's line of attention. He said the programming profession would not disappear. Every ten or fifteen years, a major technological breakthrough occurs, and each breakthrough historically increases the number of programmers.
On that view, the number will grow again now. Others predict the opposite: programmers will disappear entirely. I want to illuminate the space between those extremes because it is important to watch continuously and compare with your own profession. Someone recently wrote in the ToTheMoon comments: “I am a jeweler. When will my profession disappear? When will AI replace me?” The answer depends on what kind of jeweler you are, where you work, and how the company operates. If you work in a process-driven factory that is actively replacing workshop labor with robotics, your role may disappear quickly.
If you create unique commissioned pieces for a market that values the workmanship of one particular person, discussing near-term AI replacement may be pointless. Programming contains several especially interesting aspects. Tell us what you think. On one hand, I share the view that the number of people involved in programming will grow.
As the Yandex executive said, every new technology creates demand for more people to build and maintain it. Over recent decades, the number of programmers has only increased. Someone has to support the technology. Someone has to create millions or hundreds of millions of websites, tens or hundreds of millions of applications, and all the other software built around an endless supply of ideas. Companies and individuals continue inventing new things, markets keep moving, and products require continuous renewal.
That part is real. New technologies will create and expand many markets, and the systems will require support. Last week I recorded an episode about the singularity.
If a genuine technological singularity arrives and true artificial intelligence appears, we enter an entirely different world. For this discussion, let us set that possibility aside and examine what happens with the AI that exists now. Imagine progress stopped today. No new breakthrough arrives. We continue with systems such as ChatGPT 5.6 Sol, Anthropic Claude Fable, and Opus 5 in their different coding modes. What happens to programming and to the people who use these tools?
The number of people who understand AI will obviously grow. But will they constitute new professions, or will they simply be ordinary people using an ordinary capability? I argue constantly that they will be ordinary people. Everyone now driving a car or working in an office will eventually interact with these systems. One public estimate says seventy-five percent of programmers already use AI in development. I do not believe the global figure is that high. My estimate may be closer to twenty-five percent or less across the entire world.
The share will grow. AI will become the basis of development environments, compilers, studios, and tools everywhere, so eventually one can say that one hundred percent of programmers use it. But the important question is how they use it. While planning this episode, I thought it would be interesting to recruit several programmers who believe they understand AI well and give them tasks that I now perform myself with AI— Specifically in software development with Codex, Claude Code, and code itself.
I own and participate in very different businesses and projects. In some, I am mainly an owner and do not enter the details. In others, I still perform the role of a programmer: I sit down, build, and develop systems myself. I believe a modern person cannot develop fully without this capability. Using Codex today resembles using Windows as an operating system. Using Claude Code resembles using macOS. For an entrepreneur, business owner, marketer, salesperson, project manager, or almost any knowledge worker, refusing to use Codex or Claude Code—not merely ChatGPT as a chat window—is becoming as irrational as refusing Excel, Word, or email.
There was a period when large numbers of people and even schools did not send email or use messengers. Information was carried physically. We remember how normal that once seemed. If I work inside Codex and build something, does that make me a programmer? I was educated as a programmer and spent a substantial part of my life in software development. But I do not use these tools today merely because of that background. I use them because they represent the current movement of the world and its future.
This is a critical point to remember. Many people I know do not use Codex and do not understand why they should. I remember the same stage with email, Excel, and Word: people simply could not see the need. Even now, I meet business owners who keep financial records on paper. We often assume every entrepreneur knows the metrics of the business and maintains proper accounting. That is false. An extraordinary number—perhaps ninety percent—do not truly know their performance. The few figures they can name are elementary.
That is a different subject, however. For this episode, I am considering an experiment and would welcome ideas. We could give real work to programmers—including viewers who program—and see whether they can complete it properly from beginning to end. The difficulty became clear in a meeting I led yesterday. The participants were me, my partner and co-owner, the CEO, the head of marketing, and the
head of sales. We entered a dispute about how people should interact with AI. I told everyone something simple: I do not want to distribute tasks and then teach every person from the beginning how to work with AI. If someone neither uses nor understands these systems, it is often easier for me to complete the task myself. Teaching that person from zero no longer makes sense; I should hire someone who already understands the tools. Now consider programmers. Today Codex and Claude Code write code better than ninety-nine percent of programmers in the world.
Programmers will object. Let them object. That is my view. Across a broad range of work—testing, code generation, analysis, architecture, and systems analysis—the tools already outperform ninety-nine percent of programmers. They also produce UX and UI design better than ninety-nine percent of designers. Who constitutes the remaining one percent? Even many people in that group probably cannot compete directly with Claude Code or Codex today. But they may possess rare, unconventional knowledge that lets them create something unique even without those systems, particularly compared with someone merely operating another model.
A small group also builds and advances the AI systems themselves and understands extraordinary numbers of technical details. Let us set that group aside. The larger population of people connected with development—frontend, backend, websites, layout, simple applications, and countless other systems—numbers in the tens of millions. We will show and examine the available figures. I am recording while driving to San Francisco to meet one of my partners, and I keep checking that the microphone has not failed.
The sound may occasionally contain road noise, but you know how hard I work to record with good quality. I hope you will support the episode. Among those tens of millions are people who were programmers before the AI wave and still carry that title. Yet in the last six months alone, several million new people have almost certainly become
programmers in the broader sense. I personally know many. My sister and my editor were not programmers at all. Now they develop information and technology systems—systems that are not nearly as simple as outsiders imagine. Yesterday, for example, I began installing a new camera system. I have dozens of cameras across my property and throughout the house. The existing cameras depend heavily on cloud software and are difficult to manage, so I decided to build a different setup.
I connected directly to the new cameras through Codex and spent about thirty minutes of my own time. By this morning, the system was already reading license plates, recognizing people and animals, identifying particular vehicles, building a face collection, and giving me the option to operate both locally and through the cloud. That is real programming—and it is programming built specifically around my requirements. No ready-made product solves the whole problem because I need the camera system integrated with the smart-home systems I can access through APIs.
I recently described connecting to the control panel for my gate. The existing systems were highly isolated, so I bought a Wi-Fi device for roughly fifteen or twenty dollars. That gave me a bridge that can now communicate with the camera system. I can receive a Telegram notification when a vehicle with a particular marking arrives. What commercial system will build that exact workflow for me? None. In the past, the answer would have been to hire programmers. Now I can program a rule saying that if a squirrel remains near the property for too long, or if a skunk of a particular size appears, the system should notify me.
Someone may ask how I will train it to recognize a skunk. That is precisely the point. I provide several photographs. As the system sees animals, I mark whether each one is or is not a skunk. It learns. Open-source recognition systems already make this relatively easy. We have entered an extraordinary world. At the beginning of the year, Ilnar and I discussed a post by Andrej Karpathy, one of the foundational figures behind modern AI and the early OpenAI models. He used Codex—or what we then called OpenClaw—to connect to his cameras and build a custom solution in a few hours.
At the time, the process was difficult. Now it works. In one sense, I became a programmer again, although I had been one before. In reality, I have not built systems directly for many years. I could theoretically have created this one with conventional tools; I have enough technical background. But it would have consumed an extraordinary amount of time, and I would never have spent that time on what is partly an experiment. Some elements are genuinely useful, while others are simply entertaining.
I can make the gate open not only from a license plate, but from the plate combined with the vehicle's visual characteristics, stickers, and partial recognition of the person inside— Along with many other factors. I am interested not only in what the system can do, but in how easily it operates, how stable it is, and how easily something like it can be created. Ordinary people such as my sister and my editor have become programmers. Millions of people have become programmers.
There are not yet many users who work with Codex as deeply as they do, and capability varies widely, but the group exists. These new users are unquestionably replacing some conventional programmers. Yet my sister and editor have no intention of adopting “programmer” as their profession. They simply use these systems every day in their lives and work. A person who helps me advertise YouTube channels also uses several development systems. He has his own business and projects, partners with some of mine, and helps our team.
In the same broad sense, he too is becoming a programmer. At the same time, I personally know professional programmers who do not use AI nearly as effectively. When I discuss these systems openly, they respond, “What you are doing is not real programming,” or imply that it belongs somewhere outside their field. The camera project is only one small example among many. Someone has certainly already written in the comments that it is elementary and meaningless. Watch our other ToTheMoon videos.
Regular viewers know how many sophisticated systems we discuss and demonstrate. This camera example is a minor case. Yet some people will still dismiss it. That creates an interesting market dynamic. People such as my editor and sister do not consider themselves programmers. They would say, “What are you talking about?
I am not becoming a programmer.” Yet they are expanding into work that previously belonged to the software industry. At the same time, I would not have commissioned this particular camera system before. I would have used a ready-made product. AI therefore opens a new market, new capabilities, and new competition rather than simply replacing an existing purchase. Another group begins using Codex or Claude and explicitly calls themselves programmers even though they were not programmers before.
They sell their time, build websites for clients, and develop custom applications. I could sell a great deal of my own development work if I wanted to, but I do not sell my time that way. Theoretically I could say, “I will build this system for you,” and many people would be glad to buy it. The camera case alone could support a business in the Bay Area: designing individualized systems for homes and properties. The tools make genuinely impressive things possible— Without building a large company first.
The requirement is to think differently and earlier than the market. I open the products currently available and discover that they cannot implement what I want. The issue is not that I have an abstract desire to build another system. The existing system simply does not perform the necessary task. A partner in one of my education projects once told me: “You have finally gained the ability to create systems you could never build because your programming team always lacked the particular people able to understand you.
Now you no longer have to explain everything to them; you can implement it yourself.” He was completely right. I began developing what I personally wanted. One of my projects includes several upcoming website updates. The entire current site runs on a system built without a single traditional programmer. I and my editorial team created it. It is large and contains an enormous amount of content. For years, designers and programmers constrained me. Launching quickly and efficiently was a heavy and difficult process.
Every new system, connection, or integration required another layer of coordination. Now we can build useful and interesting integrations directly. I am recording from one of the most advanced and expensive mass-market SUVs in the world. It includes modern systems and some assisted-driving capability. It is not a Tesla—although a Tesla costs less—but I cannot connect to this Cadillac and modify its internal behavior. Cadillac has only a few ways to keep me as a customer. It can give me an extraordinary vehicle experience—materials, smell, comfort, service—or it can let me integrate with its internal systems.
Otherwise, for me personally, the product will lose. I will never build a self-driving stack from scratch, so Cadillac may lose me anyway if it cannot offer autonomous driving at Tesla's level. Tesla is pulling away from every competitor at extraordinary speed. Even Google's Waymo taxis in San Francisco lose on one dimension because they rely on lidar hardware that may cost twenty or thirty thousand dollars. That is a complex dedicated system rather than primarily software distributed through updates as in Tesla.
But let us return to the profession. People will sell work using these new capabilities, and a paradox appears: can customers identify the right person to build a software product? Today, a customer might obtain a better result by commissioning me—even though I do not sell development services—than by hiring an ordinary external programmer. The reason is that 9,999 programmers out of 10,000 may be unable to understand the complete solution. I am speaking about ordinary programming in the global market, not the rare research performed inside OpenAI or Anthropic.
Even there, my own advantage in a business problem is that I may understand the initial objective, the final outcome, and the operating logic as one system rather than seeing only a narrow service assigned for development. My editor or my sister may understand many future solutions better than a professional
programmer, precisely because they understand the domain and the actual result. What will happen in the market? More people who were never programmers will begin doing programming work. Some will be highly capable and others will be poor. Their qualification will depend not only on whether they studied computer science, but on how they think, what they know, how well they understand the subject domain, and whether they can connect the parts of a problem. Some people say programmers will become operators who manage agent systems inside Codex.
That is false. Codex, Claude Code, Grok, Meta's Muse Code and Spark, Gemini, and every other coding system are being designed to solve as much of the task independently as possible.
If the system itself performs the maximum amount of work, the central question is no longer whether a human can “manage agents” in the traditional programming sense. A new myth will say that a thousand special programmers inside Google are building unique services and therefore must possess some phenomenal programming trait. In most cases, that will not be true. Traditional programmers inside companies with thousands of engineers may face one of the market's largest problems.
They will solve a local task that once took months almost instantly, but many will be unable to understand the broader system around it. The company will then face the problem of managing ten thousand programmers whose individual tasks have collapsed in duration. The strongest opportunity may instead appear in a small manufacturing company or coffee-shop chain that previously employed one or two programmers. There, the new professional needs an unusually broad skill: not merely programming, but business analysis, systems analysis, the ability to inspect several environments, and the judgment to solve the business problem rather than execute a narrowly specified ticket.
I hear almost no serious discussion of this on the market today. The argument I am presenting is not standard. I do not say that to boast. ToTheMoon is naturally distinctive because it is made by people with unusual experience, and I believe our perspective is valuable to viewers. But my position here differs sharply from the common one. I do not think the programming, development, or systems market will disappear.
I think its structure will change. Again, we are setting aside technological singularity. If an AI eventually governs a country—or several countries—the entire environment becomes something else. Compared with that, today's blockchain debates would look like kindergarten. I am speaking even about current conditions, about what is happening now. A company with ten thousand programmers has established processes and is generally incapable of rebuilding itself. It does not understand what is really happening.
That is a disaster, because the task is not simply to replace one programmer with fifty people. The entire approach must change. Some systems are pointless to maintain; other systems need to be rebuilt from the beginning. This creates a major problem. For the last fifteen or sixteen years—since around 2010 or 2012—one of the great failures of programming has worked like this: people arrive at a company and say, “Your code was written incorrectly.” Then came endless arguments about monoliths versus independent services.
“Your code is wrong, and all of it must be rewritten.” The industry became occupied with rewriting code forever, while an enormous number of companies still operate on old code. That is a problem. It is a disaster the industry will have to confront.
I did not tell you by accident that this episode is important. It concerns every profession, not only programming. We are about to reach that broader point. I am here to give you a new way to think and to help you look differently at what is happening. My goal is not to make you run out today and change jobs, or to make five hundred times more money appear in your account tomorrow. That is not the task. ToTheMoon exists to help you understand the world in which you live, what surrounds you, and what is nearby, using technology as the lens.
All my channels are built around expanding perception. I have at least two other channels besides ToTheMoon. So what will happen? New programmers are appearing. At the same time, a huge number of existing programmers are in an unclear position. I have also watched many programmers create businesses. There are programmers who now have easier access to business management, just as businesses now have access to programming. The same cross-over is happening in every field. That structure is now spreading everywhere.
It is not spreading like water; perhaps it is closer to air. It begins to occupy every profession. This is not a discussion about programmers surviving because they will manage one particular part, or because a programmer will manage agents. That is nonsense. Will a programmer then manage marketing-employee agents?
How can a programmer automatically manage agents? Is the programmer a manager? Can a manager manage agents, or not? The entire agent discussion is a separate question. I do not particularly like the word. What happens to a salesperson when artificial intelligence gives that person an enormous number of new capabilities—assuming they use them? That produces the next question: does the person actually use these tools in their work, and how many people use them? If Codex or Claude Code remains limited to a small group instead of becoming as common as Windows or macOS, Word, or Excel, then nothing fundamental will happen.
A small number of people will possess an exclusive capability, but there will be no broad evolution. What kind of fundamental evolution could occur under those conditions? None. I believe there will be only a limited number of major systems—Codex from OpenAI and ChatGPT, Claude Code, and a few others. I have repeated this many times. If large numbers of people use them as routinely as Word, we enter a fundamentally new world. The issue is no longer that professions are being taken away.
People gain the ability to choose among more professions, while also facing extraordinarily intense competition. What happened after the collapse of the Soviet Union, or more broadly during the 1990s and 2000s? People gained greater choice and access to more professions. What happened with online education and the broad development of the internet? People gained access to more knowledge. If they wanted to do something different, they could do it and learn it faster. They no longer had to travel to a unique library in a particular place.
But that did not mean everyone chose to learn, work, or work hard. A strong trend in socially developed societies is the belief that businesspeople should rest and that people who work a great deal should switch to remote work, do less, or obtain more so-called freedom—an artificial freedom—without understanding that freedom is not defined by the number of hours you work. Freedom exists somewhere inside us. Suppose one hundred, one hundred fifty, or two hundred million people begin using Codex and Claude Code and treat connecting to a camera as casually as installing a mobile app.
For a little while longer, most people will not be able to connect to a camera the way
I did. It requires a certain kind of thinking and the knowledge that such a solution is possible. But as artificial intelligence and programming systems develop, they will perform this work for the person automatically. They may not even suggest it; they will simply do it, and the person may never know. It will be like installing a mobile application. That is why I ask: where is the programmer who supposedly sets tasks and manages the system? A different kind of thinking will win.
Do not bet on unnecessary layers. OpenAI lost an enormous amount of time and money developing a separate browser. Why build a separate external browser? It should have built the browser directly inside ChatGPT from the beginning. Even separating Codex rather than making ChatGPT itself the Codex environment cost the company time. We will see these ideas develop seriously in a different direction. The idea that Claude Code or Codex is merely a new system for programmers is nonsense.
It is a new operating system. It will look somewhat different and will be transformed, but I said this two years ago and continue saying it: ChatGPT together with all these tools is the operating system. Codex is only one component inside ChatGPT and inside the broader type of system we are discussing. If hundreds of millions of people use it as routinely as Word or Excel, then people simply
gain more choice. Programmers do not disappear. Marketers do not disappear. Advertising specialists, journalists, and editors do not disappear. Nobody disappears. People receive more choice. More choice creates more opportunity on one side and much more risk on the other. That is always the trade-off. In the Soviet Union, a person entering a university knew: “I will study here, then work in this job for thirty or forty years. This will be my life.” That was a restriction, but it removed an enormous amount of risk.
When you are given broader possibilities—you can move to this country or that city, live here, buy or rent an apartment, marry or not marry, have children or not have children—and there are no fixed rules, you face more risk than when you live according to a clearly defined structure. Wherever there is choice, there is risk. On one side there appears to be more freedom—again, often illusory freedom—and on the other
there is more risk. From the perspective of professions, therefore, the supposed problem does not exist. A real problem would arise if someone created the genuine artificial intelligence shown in many films—not like the virtual world in Ready Player One, but like the intelligence in Oblivion that begins taking over everything. That would be a problem. Under today's conditions, however, there is no such problem. There are opportunities for clear-thinking people, for people with a broad approach, for those who want to develop, and even for those whose goal is not necessarily to benefit others but who remain open to the new era in which we all live.
That is what I wish for you. Support the channel and send this video to your friends. I am confident that it will provide substantial value to a great many people. Goodbye.